DLPack integration for Rust.

This crate provides wrappers around the DLPack C API, allowing to exchange tensors with other libraries that support DLPack, such as PyTorch, TensorFlow, MXNet, CuPy, JAX, etc.

The C API is available in the sys module, and the following higher-level wrappers are provided:

• DLPackTensor: an owned DLPack tensor
• DLPackTensorRef: a borrowed DLPack tensor (immutable)
• DLPackTensorRefMut: a borrowed DLPack tensor (mutable)

We also provide tools to convert from and to other rust types, using the TryFrom trait. Conversions to and from std::vec::Vec and slices are always enabled, in the [vec] module.

Features

• sync: enable sharing data behind Arc<Mutex<_>> and Arc<RwLock<_>> with DLPack, enforcing Rust's borrowing rules at runtime through the locks.
• ndarray: enable conversion from and to ndarray::Array
• pyo3: enable passing data from and to python, following the dlpack protocol (the data is passed through a PyCapsule object).

dlpk: DLPack integration for Rust

This crate contains a direct translation of the C header for DLPack, which is intended to be used inside Rust projects that want to offer a C API.

We also provide some tools to convert to and from other types, through the following cargo features:

• ndarray to convert from and to ndarray::Array
• pyo3 to convert from and to Python data, following the python specification for dlpack

Conversions to and from std::vec::Vec and slice are always enabled.

License

This crate is distributed under the terms of both the MIT license and the Apache License (Version 2.0).